Introducing SuperAlign ScanKernel
AIRiskDB is now powered by the SuperAlign ScanKernel, a static scanner for AI assets, starting with agent skills. Add a GitHub skill repository URL and get a security read on the skill in seconds — before it ever reaches an agent.
Scanning is fully deterministic at this stage: no code execution and no model calls. Existing tooling in this space leans on LLM judgment or shallow pattern lists; SuperAlign ScanKernel is designed to improve on both, with static analysis you can rerun, audit, and trust. It will provide complete coverage for OWASP Agentic Skills Top 10 and other emerging threat frameworks for AI assets.
Scan what matters. Three scan scopes — SKILL.md only, skill folders, or the whole repository — with an optional README toggle. A review can target the instructions a skill claims, the files it would actually install, or everything sitting around them.
19 rule families, evidence included. Detections include prompt injection, data exfiltration, privilege escalation, supply chain, dangerous execution, malware signatures, MCP hygiene, and more — built on pattern analysis, structural and taint-flow code analysis, and YARA signatures. Every finding lands with a check ID, a severity, the exact file and line, and a mapping to OWASP review areas. Scan results are shareable by link.